This is a great addition. Thanks for providing so much for free
I quickly read the whole document. It seems cool but I was originally thinking that the computation of the tasks would happen on the Google App Engine but it appears that it's just the triggering of the execution of the tasks which is not as cool. Am I understanding this correctly? At some point I wanted to run an analysis of tons of web-accessible files I have and I wonder if I could have done that with this.
You can! Although there is still 30 seconds run time limit, it's much more better than cron based solutions. For one you can run tasks in parallel.
You can also use something like schema-migration example from Google IO presentation. So one task executes and after it's done adds new task to task queue.
They also plan MapReduce on top of task queue, that'll be even better suited for your job.